How to use a Venn diagram in probability

The four regions of a two-event Venn diagram

A Venn diagram is a picture of a sample space. The rectangle around the outside holds every possible outcome, and each circle holds the outcomes in one event. With two events $A$ and $B$ the picture splits into at most four regions, and every outcome lands in one and only one of them:

• $A$ only: in $A$, not in $B$.
• The overlap: in both $A$ and $B$.
• $B$ only: in $B$, not in $A$.
• Neither: inside the rectangle but outside both circles.

If the events cannot both happen, the overlap region is empty and only three regions carry outcomes.

Those probabilities add to 1, because the regions cover the whole sample space without overlapping. That is true by construction rather than by luck, so treat it as the fact the picture is built on and not as a test that will catch your errors. The tests that catch errors come later: no region may come out negative, and each circle must still add back to the total you were given.

The reason a Venn diagram is worth drawing is that the phrases in a probability question map onto regions. "At least one of them" is three regions. "Exactly one of them" is two. "Neither" is one. Once the four numbers are written into the picture, most questions become addition.

Where union, intersection, and complement sit

Three symbols name the regions, and each has an everyday translation.

Two details cost points every year. First, "or" in statistics is inclusive: $A \cup B$ includes the overlap, so "A or B" means at least one of them, not one but not the other. Second, the complement is taken against the rectangle, not against the other circle, so $A^c$ contains the $B$-only region and the neither region.

The region outside both circles has a name worth knowing: it is $(A \cup B)^c$, the complement of the union. So $P(\text{neither}) = 1 - P(A \cup B)$, which is usually the fastest way to get it. If you want the full comparison of the first two symbols, see union vs intersection in probability.

Reading the addition rule off the picture

Here is the whole idea. Circle $A$ covers the $A$-only region plus the overlap. Circle $B$ covers the $B$-only region plus the overlap. Add $P(A)$ and $P(B)$ and the overlap has been counted twice, once from each circle. Subtract it once and it has been counted once, which is what you wanted:

The sign is negative, and there is a fast way to be sure. Probabilities cannot exceed 1. In the class example below, the sport event has $P(S) = \frac{18}{30} = 0.6$, the instrument event has $P(I) = \frac{12}{30} = 0.4$, and the overlap is $\frac{7}{30} \approx 0.2333$. Adding all three gives $1.2333$, which is impossible, so the overlap term cannot carry a plus sign. That argument runs one way only. A formula that produces a probability above 1 on real numbers is certainly wrong, but a result under 1 does not certify the signs, because with small probabilities a wrong arrangement can still land inside the legal range.

When the two events are mutually exclusive the circles do not touch, so $P(A \cap B) = 0$ and the rule collapses to $P(A \cup B) = P(A) + P(B)$. That is the only situation where you may simply add. Rolling one fair die, "even" and "shows a 5" cannot both happen, so $P(\text{even or } 5) = \frac{3}{6} + \frac{1}{6} = \frac{4}{6}$, and counting the outcomes 2, 4, 5, 6 straight off the diagram gives the same $\frac{4}{6}$.

Fill the overlap first, then work outward

Questions almost never hand you the four region probabilities. They hand you $P(A)$, $P(B)$, and $P(A \cap B)$, which describe circles, not regions. Convert in this order.

1. Write the overlap, $P(A \cap B)$, in the lens.
2. $A$ only $= P(A) - P(A \cap B)$. Subtracting is what keeps the overlap from being counted twice.
3. $B$ only $= P(B) - P(A \cap B)$.
4. Neither $= 1 - (\text{the three regions you just wrote})$.

Step 4 cannot fail on its own, because computing the outside region as the leftover forces the four to sum to 1 no matter what the first three were. The checks that do bite are these: every region must come out non-negative, and adding the overlap back to each single region must return the circle totals you were given. A negative outside region means $P(A)$, $P(B)$, or the overlap was misread. Work in counts rather than probabilities whenever the problem gives you counts, since whole numbers are easier to check, and divide by the total at the end.

A diagram filled in this way answers the awkward phrasings directly. "Exactly one" is the two single regions added, which also equals $P(A \cup B) - P(A \cap B)$. "At least one" is the union. "Neither" is the outside region. "Not both" is everything except the lens, or $1 - P(A \cap B)$.

Three events, and the signs people reverse

With three circles the same double-counting logic runs one level deeper, and this is where sign errors live:

Add the three single events, subtract the three pairwise overlaps, then add the triple overlap back. The pattern alternates plus, minus, plus, and each step has a reason. The three pairwise overlaps were each counted twice by the singles, so subtracting fixes them. The center region was counted three times by the singles and then subtracted three times by the pairs, leaving it counted zero times, so it has to be added back once.

Test it on numbers before you trust it. In the streaming survey worked below, the correct signs give $0.47 + 0.37 + 0.28 - 0.17 - 0.13 - 0.11 + 0.05 = 0.76$, which matches the 76 households counted directly off the regions. Flip the signs to plus, plus, minus and you get $1.48$, a probability greater than 1.

Here the impossible answer exposes the mistake, but it will not always. Shrink the numbers and the wrong arrangement stays legal: with $P(A) = P(B) = P(C) = 0.1$, each pairwise overlap $0.02$ and the triple overlap $0.01$, the correct signs give $0.25$ while the reversed ones give $0.35$, and nothing about $0.35$ looks wrong. Going above 1 is a tell, not a proof. The check that always works is the picture: count the union off the regions and compare it with what the formula returned.

Conditional probability and independence from the same picture

A Venn diagram also carries conditional probability, once you notice what conditioning does: it throws away the rectangle and makes one circle the new whole.

Read "$P(B \mid A)$" as "the probability of B given A". On the diagram, you are standing inside circle $A$ and asking what fraction of it is shaded by $B$. That fraction is the lens divided by the whole of circle $A$. Because the denominators differ, $P(B \mid A)$ and $P(A \mid B)$ are usually different numbers, which is the mix-up covered in conditional probability and why order matters.

Independence is a numerical check, not a picture. Events are independent when $P(A \cap B) = P(A)P(B)$. In the class example, $P(S)P(I) = 0.6 \times 0.4 = 0.24$ while the actual overlap $P(S \cap I)$ is about $0.2333$, so those two events are close to independent but not independent. Do not read overlap as dependence and separation as independence: two circles that do not touch are mutually exclusive, which forces the events to be dependent whenever both have nonzero probability. Disjoint vs independent events is worth reading once, carefully.

Mistakes to avoid

• Writing $P(A)$ into the $A$-only region. $P(A)$ describes the whole circle, so the $A$-only region is $P(A) - P(A \cap B)$.
• Adding probabilities without subtracting the overlap. That is legal only when the events are mutually exclusive.
• Reading "or" as exclusive. $A \cup B$ includes outcomes in both.
• Forgetting the neither region, then wondering why the four numbers do not reach 1.
• Reversing the three-event signs. Plus, minus, plus. If your union comes out above 1, or above the largest possible count on the diagram, you have found the error.
• Treating a Venn diagram as a probability engine for repeated trials. It shows one sample space at one moment. For staged, sequential events use a tree diagram instead, and for repeated success or failure trials start from a Bernoulli trial.